自适应精度保留的浮点数时间序列无损压缩算法研究

Existing lossless compression schemes for floating-point time series are fundamentally bifurcated into two methodological paradigms: XOR-oriented techniques and decimal encoding strategies. XOR-based mechanisms mitigate redundancy through bitwise XOR operations applied to successive IEEE 754 floating-point values, utilizing the recurrence patterns of leading-zero lengths and central significant bits. Nevertheless, these techniques demonstrate performance degradation when processing numerically proximate values exhibiting substantial binary representation discrepancies under the IEEE 754 standard, yielding limited practical redundancy exploitation. Such scenarios frequently result in suboptimal compression ratios or even compression inefficacy. Decimal encoding methodologies transform binary floating-point numbers into integer representations via 10^n scaling, though divergent decimal precision requirements typically entail amplified storage overheads. This paper proposes an Adaptive Precision Preservation framework for Lossless Compression of Floating-Point Time Series (APCF). The APCF methodology comprises two principal operational phases: preprocessing and encoding. During the preprocessing phase, the algorithm adaptively identifies optimal decimal precision thresholds, performs adaptive quantization, and consolidates residual information in least significant bits through bitwise XOR transformations. The encoding phase employs run-length encoding (RLE) for precision metadata compression coupled with a modified Gorilla encoder for residual processing. Extensive evaluations across 18 temporal datasets demonstrate APCF\'s superiority, yielding a mean compression ratio of 0.23:1. Comparative analysis with state-of-the-art compressors ALP and ELF reveals relative compression gains of 18.2% and 19.9% respectively. The source code repository is accessible at https://github.com/liuyuxicici/APCF.git.